Comparing the Accuracy of Reconstructed Image Size in Super

Jul 30, 2015 - Another way to visualize the apparent dsDNA locations is to create spatial frequency histograms, Figure 3C and D. The data from the sca...
1 downloads 7 Views 5MB Size
Download PDF
Article pubs.acs.org/JPCC

Comparing the Accuracy of Reconstructed Image Size in SuperResolution Imaging of Fluorophore-Labeled Gold Nanorods Using Different Fit Models Karole L. Blythe,† Eric J. Titus,† and Katherine A. Willets*,†,‡ †

Department of Chemistry, The University of Texas at Austin, 102 East 24th Street, Austin, Texas 78712, United States Department of Chemistry, Temple University, 1901 North 13th Street, Philadelphia, Pennsylvania 19122, United States

‡

S Supporting Information *

ABSTRACT: We use a triplet-state-mediated super-resolution fluorescence imaging technique to localize the position of individual fluorescently labeled double-stranded DNA (dsDNA) bound to the surface of gold nanorods. Within each diffractionlimited spot, we must account for two different emission sources: the stochastic fluorescence from the fluorescent labels and the steady background luminescence of the gold nanorod. To isolate the contribution from the fluorescent label, we subtract the average gold nanorod luminescence contribution, modeled with either a two-dimensional Gaussian or a dipolar emission model. The fluorescence from the labeled dsDNA is then fit with a two-dimensional Gaussian to reconstruct the positions of each individual emitter on the nanorod surface. The resulting reconstructed images, using either luminescence model, agree with the shape and orientation of the underlying nanorod, and show similar apparent dsDNA binding heterogeneity across the surface of the nanorod based on the localization of the fluorescent labels. Using the dipolar emission model for the luminescence allows for the retention of more emission events from the fluorescent label, after applying a fitting threshold, and yields a more robust reconstructed image containing more centroid points that show the apparent locations of the dsDNA. Unfortunately, the sizes of the reconstructed nanorod images were smaller than expected, despite the use of the more accurate model for the gold luminescence, suggesting that the photophysics of this coupled dye−nanorod system are more complicated than when using the isolated fluorophores.

■

resolution fluorescence imaging techniques to map the location of fluorescently labeled double-stranded DNA (dsDNA) bound to the surface of single AuNRs via a thiol linker.18 While we were able to recover the shape and orientation of the AuNR support in the reconstructed images, we found a consistent underestimation in the size of the nanorods. Similar disagreements have been observed by Uji-i and co-workers, who reported differences between the actual diameters of silver nanowires and the reconstructed image size produced by superresolution imaging of fluorescent proteins tethered to the nanowire surface.19,20 In their case, they attributed the size mismatches to distortions of the point spread function (PSF) of the fluorescent emitter, introduced by its proximity to the metallic substrate.19 These PSF aberrations were apparent by eye in their raw data, and showed a dependence on the width of the nanowires studied (110 and 250 nm).19 In our previous work, and in the work presented here, we have used both nanowires and nanorods with widths below 100 nm (∼68 and ∼35 nm, respectively) and have seen no visible evidence of PSF

INTRODUCTION Functionalizing gold nanoparticles with different ligands has become a popular technique for increasing the versatility of how these nanoparticles can be used, in applications such as bioimaging,1−5 drug delivery,4,6,7 photothermal therapy,2,5,8,9 and surface-enhanced spectroscopy.3,10,11 Gold nanorods (AuNRs) are often used as the substrate for these experiments due to their easily tunable localized surface plasmon resonance,8,12,13 photostability,14,15 and local heating properties.8,16 While it is important to understand how ligands assemble on the AuNRs to optimize the functionalization chemistry for specific applications, bulk measurements cannot provide information about how particle surface heterogeneity can affect ligand binding and performance. Recently, super-resolution imaging has emerged as a tool for studying the position of single fluorescent molecules tethered to a wide range of substrates, including the cytoskeleton in eukaryotic cells, cellular membranes, and metallic nanostructures.17−24 In many of these studies, the shapes and sizes of the underlying substrates can be reconstructed, often with high fidelity, by mapping the positions of each fluorescent tag on the substrate surface and building a composite image from the individual emitters. Previously, we have used similar super© 2015 American Chemical Society

Received: May 25, 2015 Revised: July 23, 2015 Published: July 30, 2015 19333

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

Article

The Journal of Physical Chemistry C distortions in our raw data;17,18 thus we seek an alternative explanation for the lower-than-expected dimensions of the reconstructed AuNR super-resolution images. In this Article, we explore the possibility that the choice of fitting model in the super-resolution analysis impacts the size of the reconstructed images. For our super-resolution studies, we use carboxytetramethyl rhodamine (TAMRA) as a fluorescent label attached to dsDNA ligands. TAMRA was chosen because its fluorescence maximum, ∼580 nm, does not overlap with the localized surface plasmon resonance of the AuNRs (640−690 nm for the functionalized AuNRs discussed in this Article, Figure S1) based on single particle dark field microscopy scattering spectra. This combination of fluorophore and nanorod was chosen to lower the probability of TAMRA emission coupling into the AuNR, which leads to surface enhanced fluorescence25 and potentially skewed positions of the fluorophore.26 TAMRA molecules can be switched between a fluorescent (on) and nonfluorescent (off) state by promoting intersystem crossing into nonemissive triplet states, a technique known as groundstate depletion with individual molecule return (GSDIM).17,18,23,27 DNA is an ideal ligand for these studies because it is thought to assemble into highly ordered monolayers on the nanoparticle surface, spacing the TAMRA molecules a fixed distance from the surface.28−30 Ideally, only one TAMRA molecule is on at a time and its diffraction-limited emission is fit to a two-dimensional (2-D) Gaussian eq 1, where I(x,y) is the spatially dependent intensity, z0 is the background intensity, I0 is the peak intensity, sx and sy are the widths of the Gaussian in x and y, and x0 and y0 are the Gaussian center-ofmass (hereafter referred to as the centroid). The location of the single TAMRA emitter is approximated as the centroid position (x0, y0). The process is then repeated multiple times, such that the position of each TAMRA molecule is localized, to gain a complete picture of where the fluorescently labeled dsDNA ligands are bound to the AuNR surface. 2

I(x , y) = z 0 + I0 e[−1/2[((x − x0)/ sx)

+ ((y − y0 )/ sy)2 ]]

bound to the surface, which is information that is hidden in a bulk measurement.18 However, while the images reconstructed from the super-resolution data matched the shape and orientation of the underlying AuNR, the size of the reconstructed image was always smaller than the expected dimensions of the AuNRs. If we were mapping the true location of the dsDNA, then we would expect the reconstructed images to match the measured dimensions of the AuNR, assuming uniform coverage of dsDNA. We hypothesized that our choice of a 2-D Gaussian for fitting the AuNR luminescence could be influencing the super-resolution results. This model was chosen for its simplicity and because it has been used by the bulk of the super-resolution community, yet it has no physical meaning and can introduce significant aberrations into the fits.24,31−33 On the other hand, our group has previously shown that AuNR luminescence is well-modeled as the sum of three mutually orthogonal emitting dipoles at an interface; we will refer to this as the 3-dipole model through the rest of this Article.31,34 In the 3-dipole model, a dipole corresponding to the longitudinal axis of the AuNR is described by the in-plane angle of the AuNR along its long axis, ϕ, and the out-of plane angle of the AuNR, θ (see Figure S2 for angle definitions). Two other mutually orthogonal dipoles are included to model the short axes of the AuNR. The emission from the three dipoles sitting at the air-glass interface is propagated through the glass coverslip substrate and the associated optics, and then projected onto an imaging camera with finite pixel size to generate a calculated PSF of the AuNR luminescence. Unfortunately, this model does not have a closed form expression and must be evaluated numerically. More details on this model can be found in the Experimental Methods. In this Article, we compare the results of mapping the location of ligands on the surface of AuNRs when the AuNR luminescence is fit with either the 3-dipole model or a 2-D Gaussian. The dsDNA labeled with a TAMRA fluorescent reporter molecule remains the ligand in these experiments, and we continue to use the triplet-state-mediated technique to photoswitch the TAMRA molecules for our super-resolution experiments.17,18 Although the 3-dipole model is more computationally expensive than the 2-D Gaussian due to the lack of closed form expression and therefore increased computation time,31 we expect it to yield more accurate results for localizing ligands on the surface of AuNRs.

(1)

Unfortunately, the underlying AuNR substrate is inherently luminescent and produces a constant intensity contribution to the diffraction-limited emission. Because the AuNR and TAMRA dyes are spaced by ∼10 nm, the two emitters cannot be spatially resolved, according to the Abbe criterion. Failing to account for the presence of the AuNR background leads to a calculated centroid that reflects the intensity-weighted superposition of both the TAMRA and the AuNR emission, generating incorrect information about the position of the dsDNA ligands on the surface. Therefore, we need a method to differentiate between the positions of the two signals: (1) bound TAMRA molecules undergoing photoswitching and (2) steady AuNR luminescence, to more accurately determine the position of each ligand on the surface. In previous work mapping the position of fluorescently labeled dsDNA attached to a AuNR, we fit the AuNR signal (IAuNR) using eq 1, and then subtracted the results of that fit (IAuNR,fit) from frames where both TAMRA and the AuNR luminescence were present (Icombined) to isolate the contribution from the TAMRA alone (ITAMRA = Icombined − IAuNR,fit).18 We then fit ITAMRA to eq 1 to localize the centroid position of the fluorophore and thus the apparent position of the dsDNA ligand. We found significant heterogeneity among different AuNRs with respect to the apparent locations of the dsDNA

■

EXPERIMENTAL METHODS AuNR Synthesis. Cetyltrimethylammonium bromide (CTAB) coated AuNRs were synthesized using a seedmediated technique.18,35,36 All chemicals were purchased from Sigma-Aldrich, and 18.2 MΩ cm resistivity nanopure water was used to make solutions. Briefly, 0.25 mL of 0.01 M chloroauric acid (HAuCl4), 9.75 mL of 0.1 M CTAB, and 0.6 mL of 0.01 M ice cold sodium borohydride (NaBH4) were combined in a scintillation vial to make the seed solution. The solution sat at room temperature for 2 h to allow for seed formation to come to completion. Next, the growth solution was prepared by mixing 40 mL of 0.1 M CTAB, 2 mL of 0.01 M HAuCl4, 0.6 mL of 0.01 M silver nitrate (AgNO3), 0.8 mL of 1 M hydrochloric acid (HCl), and 0.32 mL of 0.1 M ascorbic acid into an Erlenmeyer flask. Finally, 100 μL of 10% seed solution, diluted with nanopure water, was added to the growth solution. The solution sat at room temperature overnight to ensure the completion of AuNR growth. 19334

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

Article

The Journal of Physical Chemistry C AuNR Overgrowth. Next, the AuNRs were coated with a thin layer of Au to cover any silver atoms on the surface of the AuNR to provide a better surface for binding thiolated dsDNA to the nanoparticle.18,28 4 mL of AuNRs was centrifuged for 20 min at 10 000 rpm and resuspended in 4 mL of nanopure water. The approximate concentration of AuNRs in the 4 mL solution was calculated using the bulk extinction spectrum, taken with a UV−vis, and the extinction coefficient of 4.4 × 109 M−1 cm−1.18,37 The AuNRs were centrifuged again and resuspended in sufficient 0.01 M CTAB to reach a AuNR concentration of ∼900 pM. Next, ascorbic acid and HAuCl4 was added to bring the concentrations in the solution to 1 and 0.005 mM, respectively. The solution was allowed to sit at room temperature for 1 h, after which CTAB was added to bring its concentration to 0.05 M. AuNR Functionalization with dsDNA Labeled with TAMRA. This protocol is described in detail in previous work published by our group and is based on the protocol published by Mirkin and co-workers.18,28 The first step of the AuNR functionalization protocol involves hybridizing thiolated singlestrand DNA (ssDNA) to its complement to form the dsDNA. For these experiments, we used a mixture of TAMRA-labeled ssDNA and unlabeled ssDNA complements such that the resulting dsDNA was thiolated on one end, and was either labeled or unlabeled on the other. The unlabeled dsDNA was introduced as a spacer to avoid self-quenching from nearby TAMRA molecules on the surface and to provide structural support to hinder the ability of the labeled dsDNA to bend or flop toward the AuNR surface. Ideally, the unlabeled dsDNA has an equal probability of binding to the AuNR surface as the labeled dsDNA. The ssDNA strands were purchased from Integrated DNA technologies. For the labeled dsDNA, we hybridized thiolated single strands (5′AAGAATTTATAAGCAGAAAAAAAAAAAA[thiol]-3′) and TAMRA labeled single strands (3′-[TAMRA]TTCTTAAATATTCGTCTTTTTTTTTTTT-5′). The unlabeled dsDNA was made by hybridizing thiolated single strands with unlabeled single strands (3′TTCTTAAATATTCGTCTTTTTTTTTTTT-5′). The labeled dsDNA and unlabeled dsDNA were hybridized in separate Eppendorf tubes by mixing 25 μL of the respective strands (50 μL volume total). The Eppendorf tubes were put in a 95 °C water bath for 2 min and then were allowed to cool at room temperature for 1 h. The dithiol on the thiolated ssDNA is cleaved by adding 100 mM dithiothreitol (DTT) to each of the Eppendorf tubes. After 30 min, the dsDNA was filtered using a Centri-Spin column (Princeton Separations). 100 μL of the overgrown AuNRs was centrifuged twice for 20 min at 10 000 rpm and was resuspended in 50 μL of labeled dsDNA and 50 μL of unlabeled dsDNA after the second centrifugation. This solution was allowed to sit at room temperature for 1 h. The final concentrations of labeled dsDNA and unlabeled dsDNA were 7.2 nM and 7.2 μM, respectively, in the 100 μL solution, which means that solution stoichiometry predicts a 1:1000 ratio between labeled and unlabeled dsDNA on the surface. After 1 h, 1 μL of 1% sodium dodecyl sulfate (SDS) and 10 μL of 0.01 M phosphate buffer were added and the solution was allowed to sit for 30 min. Next, the sodium chloride (NaCl) concentration was gradually increased, by adding varying amounts of 1 M NaCl for electrostatic screening of the dsDNA. The amounts (6, 6, 12, 14, 15, and 15 μL) were added in 30 min intervals. The solution sat overnight and was

centrifuged three times (20 min for 10 000 rpm) and resuspended in 100 μL of 0.01% SDS each time. After the final resuspension, the solution was placed in the refrigerator for storage. Coverslip Preparation. 25 × 25 mm #1 thickness glass coverslips were used in these experiments. The coverslips were patterned with an aluminum alpha-numerical grid via shadow deposition.38 The patterned coverslips were first cleaned with argon plasma (Harrack plasma cleaner) for 15 min. The coverslips were then rinsed with ethanol and nanopure water, dried under nitrogen, and placed in a 10 mL solution of 0.5% (3-aminopropyl) triethoxysilane (APTES) in ethanol, which was shaken on a rocker for 2 min. The coverslips were then rinsed with ethanol and nanopure water and dried under nitrogen. Five microliters of 20:100 dilution of the functionalized AuNRs in 0.01% SDS was drop cast onto the slide, and the drop was allowed to sit for 1 min so the AuNRs could adhere to the coverslip surface. The coverslip was then rinsed with nanopure water, to remove unbound functionalized AuNRs, and dried under nitrogen. Finally, 5 μL of 1:50 dilution of sky blue fluorescent beads (Spherotech) in water was drop cast onto the slide. The drop sat for 45 s, and then the slide was rinsed with nanopure water and dried under nitrogen. Optical Microscopy. The functionalized AuNRs were imaged via epi-illumination using an inverted Olympus microscope (IX-71) with an Olympus 100× oil-immersion objective that has an internal iris allowing for variable numerical apertures between 0.6 and 1.3 (UPLFLN100XOI). We used the 1.3 numerical aperture for the fluorescence experiments. The excitation source, a 532 nm CrystaLaser, was passed through a quarter wave plate to achieve quasi-circularly polarized light. The light was then passed through a lens at the back of the microscope for widefield illumination at the sample plane. Next, the beam was reflected off a 532 nm dichroic beamsplitter (Semrock, Di02-R532-25x36) and passed through the objective to be incident on the sample. The laser intensity at the sample plane was ∼29 kW/cm2. This laser intensity is sufficiently high to promote high populations of TAMRA to the triplet state, while preventing AuNR damage, as verified by dark field scattering. TAMRA fluorescence and AuNR luminescence were collected back through the objective, passed through a long pass filter (Semrock, LP03-532RU), and imaged on an EMCCD (Princeton Instrument, PhotonMax). 3650 frames (356 × 356 pixels) with an exposure time of 33 ms/frame were taken at a time. Multiple 3650 frame movies were taken to increase the likelihood of probing all of the TAMRA molecules that were attached to the AuNR surface. The coverslip was covered with a home-built nitrogen chamber, and all of the optical experiments were done under a continuous flow of nitrogen. Atomic Force Microscopy (AFM). Correlated AFM images of the functionalized AuNRs that were imaged on the optical microscope were also obtained.39 We used the alphanumerical grid on the coverslip to locate the same functionalized AuNRs between the two systems, using a combined NTMDT NTegra Vita AFM and total-internal reflection optical microscope for the structure correlation AFM imaging. Data Processing. The data consist of *.tiff stacks with 3649 frames per movie and are processed using in-house written MATLAB code. The data analysis process is described in detail in previous work and will only be briefly summarized here.18 The first frame of each movie is discarded due to timing errors of the CCD camera. To analyze the data, first we differentiate 19335

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

Article

The Journal of Physical Chemistry C between “off” frames (AuNR luminescence only, IAuNR) and “on” frames (TAMRA emission plus AuNR luminescence, Icombined) by subtracting adjacent frames. If signal remains in the subtracted image, then this is initially identified as an “on” event; otherwise, the event is initially labeled as an “off” event. Next, the emission intensities from all “off” events, which should be associated with AuNR luminescence only, are placed into a histogram; the intensity is expected to be normally distributed around some mean value if only AuNR luminescence is present. However, we find significant high intensity outliers in this step, indicating that several “on” frames have been misassigned. To reject these frames, we only keep “off” frames that are a single standard deviation above the mean luminescence intensity. Even after this rejection step, we find the luminescence intensity data are skewed from a normal distribution toward higher intensity values. We histogram these data and fit them to a Gumbel distribution, then further reject any “off” frames that are one standard deviation above the mean of the fit to this distribution. This strategy ensures that we are only assigning “off” frames associated with AuNR luminescence and are not biased by images in which a weak TAMRA emitter is also present. Next, we identify “on” events as those that are five standard deviations above the mean AuNR luminescence. Although this strategy will reject weak TAMRA emitters, those events have sufficiently low signal-to-noise that they will not be well-fit to a 2-D Gaussian for the superlocalization analysis and would be rejected later in the data processing. The “off” frames are then fit with either the 3-dipole model or 2-D Gaussian to calculate the average AuNR intensity and position, allowing us to reconstruct the contribution of the AuNR to each image; this will be referred to as the average AuNR contribution (IAuNR,fit,avg). For the case of the 2-D Gaussian model, we fit every “off” frame and take the average of all fits; however, for the 3-dipole model, we randomly select 5 frames to be fit, to reduce the computation time (each frame takes ∼20 min to fit with this model). The 5 frames are selected from differing times during the movie, to reduce any potential time-dependence in the results. The average AuNR contribution is corrected for stage drift (as described below), then subtracted from each “on” frame, leaving only TAMRA emission (ITAMRA = Icombined − IAuNR,fit,avg). The remaining TAMRA emission is fit with a 2-D Gaussian using a bounded nonlinear least-squares method to obtain its centroid position with respect to the AuNR surface. Only fits with R2 values above 0.8 are kept. While this value is somewhat arbitrary, it ensures that fits that do not well model the data are excluded from the final reconstructed images (as will be seen in Figures 1 and 2). As with other single molecule reports in the literature, we cannot definitively confirm that the signal originates from only a single emitter, although the digital stepwise nature of each emission event (as seen in the time traces shown in Figure 2) strongly suggests single molecule behavior. The series of movies that are collected for each functionalized AuNR are processed separately. We use the average AuNR luminescence position as a reference point to compile all of the TAMRA centroid positions that were localized for each movie to produce the final scatter plot. To account for mechanical drift over the course of the experiments, we use fluorescent sky blue beads (Spherotech) as alignment markers.40 Regions must have at least two alignment markers to be analyzed. The emission from each sky blue bead within the region of interest is fit to eq 1, and a drift correction

Figure 1. (A) Optical image of AuNR luminescence. (B) Correlated AFM image of the AuNR shown in (A). (C) 3-Dipole fit to the AuNR luminescence image. (D) 3-Dipole fit residuals (image − fit). (E) 2-D Gaussian fit to the AuNR optical image. (F) 2-D Gaussian fit residuals (image − fit). Images A, C, D, E, and F share a common 500 nm scale bar. Image B has a 50 nm scale bar.

is calculated for each frame by subtracting the centroid position of each bead at frame N from its centroid position at frame 2 (recall: frame 1 is discarded due to timing issues). The individual correction trajectories are compared by eye to ensure that individual beads are following identical drift trajectories; otherwise, the region of interest is rejected. The drift tends to move in a single general direction within a single movie, although the direction of the drift will vary from experiment-toexperiment. The corrections for each bead are then averaged to yield a correction trajectory over all frames of the movie, which is then used to correct for drift in the calculated AuNR and TAMRA positions. 3-Dipole Model. As described above, the AuNR luminescence is described as the sum of three mutually orthogonal dipoles at an air−glass interface. The primary dipole is oriented along the long axis of the AuNR with in-plane orientation angle, ϕ, and the out-of plane angle, θ (Figure S2), and the other two dipoles are defined to be mutually orthogonal to the primary dipole. To model the dipole emission, we have 10 variables that are fit: the defocus of the microscope, the wavelength of emission (λ), ϕ, θ, the peak luminescence intensity, the background, the position of the AuNR emission (xNR, yNR), and the contribution of each dipole component to the total emission intensity.31 The emission wavelength, λ, is fit to the same value for all three dipole components, based upon previous work, which showed that allowing each dipole to have a unique value of λ (e.g., a fit with 12 variables) yielded nonphysical results. In addition to the variables that are fit during our data processing, we must also include experimental parameters such as the numerical aperture of the microscope objective; the magnification of the microscope; the refractive indices of the glass coverslip, microscope immersion oil, and imaging medium; and the distance of the AuNR from the surface (set at zero based on 19336

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

Article

The Journal of Physical Chemistry C

average, it takes ∼1 s/frame to fit AuNR luminescence to a 2D Gaussian and 18 min/frame when using the 3-dipole model, indicating the computational expense of using this more physically rigorous model.31

■

RESULTS AND DISCUSSION Figure 1A shows the optical image of AuNR luminescence only (an “off” frame). The AuNR emission pattern indicates dipolelike emission of the AuNR, which can be seen in the bright central feature and two oriented side lobes.22,31,39,39,41−44 The side lobes, which are perpendicular to the long axis of the AuNR, provide a visual aid to assess the orientation of the long axis of the AuNR.22,39,43,45−47 The AuNR images in Figure 1 have been contrasted to highlight the side lobes, leading to image saturation of the bright central feature; the unsaturated version of Figure 1 can be found in Figure S3. We included the correlated AFM image of this AuNR, Figure 1B, to show that we are imaging a single AuNR and that the orientation of the AuNR in the AFM matches the orientation of the side lobes in the optical image. The 3-dipole fit for this off frame is shown in Figure 1C. The fit was able to accurately model the bright central feature as well as the two oriented side lobes. To further assess the accuracy of the 3-dipole fit, we calculate fit residuals (image − fit). The residuals in Figure 1D are random and have low intensity relative to the intensity of the original image. This indicates that the 3-dipole model correctly models the original optical image. Figure 1E shows the 2-D Gaussian fit to the AuNR optical data, which is only able to model the central feature. The 2-D Gaussian fit residuals, Figure 1F, are nonrandom with high relative intensities in comparison to the 3-dipole fit, indicating that the 2-D Gaussian is a poor model for this emitter. Although we could use a rotated Gaussian to allow the emission to be to described as an ellipse, we use the 2-D Gaussian shown in eq 1 throughout the remainder of this work for exact comparison to our previous work.18 Figure 2 depicts how the choice of AuNR luminescence model can affect the TAMRA emission contribution (ITAMRA) that is calculated after subtracting the average AuNR luminescence. Figure 2A shows a representative integrated intensity versus frame (time) trace of the functionalized AuNR from Figure 1. When imaging CTAB-capped AuNRs or AuNRs functionalized with only unlabeled-dsDNA, we observe constant AuNR luminescence (vide infra); therefore, we know that the intensity bursts correspond to the TAMRA molecules undergoing GSDIM. The red dots indicate frames that our algorithm identifies as “on” (Icombined = ITAMRA + IAuNR), and the green dots indicate “off” frames (AuNR luminescence only). We include a zoomed-in portion of the time trace, inset in Figure 2A, to provide a clear view of the steady AuNR luminescence background and the TAMRA emission intensity bursts above that background level (note that some frames are not marked as either “on” [red dot] or “off” [green dot], indicating that the signal did not allow us to make a clear assignment; these frames are not included in the analysis). The optical image shown in the inset corresponds to frame 2487 of this movie and is assigned as an “on” frame. We subtracted the calculated average AuNR contribution (IAUNR,fit,avg) from the image in Figure 2A, using either the 3-dipole or 2-D Gaussian model, to see how the remaining TAMRA contribution (ITAMRA) was impacted by the choice of model. Figure 2B is the average AuNR luminescence contribution, as fit with the 3dipole model. After subtracting away the 3-dipole modeled

Figure 2. (A) Representative integrated intensity versus frame number (time) trace for a TAMRA-labeled AuNR. The red dots signify frames identified as TAMRA emission events (“on” events), and the green dots signify AuNR luminescence only (“off” events). The inset shows a zoomed-in portion of the trace, as well as the optical image for an “on” event at frame 2487. (B) Average AuNR luminescence contribution, calculated with the 3-dipole model. (C) TAMRA emission contribution to this on event (ITAMRA = Icombined − IAuNR,fit,avg), where the AuNR luminescence is fit with the 3-dipole model and subtracted from the combined image. (D) 2-D Gaussian fit and (E) fit residuals for the subtracted image from (C). (F) Average AuNR luminescence contribution, calculated with the 2-D Gaussian model. (G) TAMRA emission contribution to this on event (ITAMRA = Icombined − IAuNR,fit,avg), where the AuNR luminescence is fit with the 2-D Gaussian model and subtracted from the combined image. (H) 2-D Gaussian fit and (I) fit residuals for the subtracted image from (G). Images B−I and the optical image in A share a common 500 nm scale bar.

previous work).31 Initial guesses are put into our fitting code based upon visual inspection, and then a bounded nonlinear least-squares minimization is used to fit each “off” frame. Outputs from the fit to a single frame are then used as initial guesses for the next frame to reduce computation time. In previous work, we performed this analysis on 59 AuNRs and showed that this approach yields output parameters (ϕ, θ, λ) that agree extremely well with their experimentally measured values.31 The results from previous data also showed that, on 19337

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

Article

The Journal of Physical Chemistry C average AuNR contribution from the original image in Figure 2A, we are left with the contribution from TAMRA emission (ITAMRA = Icombined − IAuNR,fit,avg, Figure 2C); we will refer to this as the subtracted image. The subtracted image is then fit with a 2-D Gaussian, Figure 2D, to extract the centroid position of the reporter molecule. The residuals in Figure 2E represent the subtracted image minus its 2-D Gaussian fit. These fit residuals are fairly random and have a low relative intensity, indicating that a 2-D Gaussian was able to model the remaining TAMRA emission after the 3-dipole-modeled AuNR contribution was subtracted from the original image. The TAMRA emission fit has an R2 value of 0.85, which passes our R2 threshold of 0.8, so its centroid position would be included in the final reconstructed image for this functionalized AuNR. For comparison, we also fit the average AuNR luminescence contribution with a 2-D Gaussian, Figure 2F, which does not accurately capture the emission features of the AuNR, as shown in Figure 1. This can be seen in the subtracted image, Figure 2G, where the two side lobes from the AuNR luminescence emission pattern remain apparent. When the subtracted image (Figure 2G), which should only contain the contribution from TAMRA emission, is fit with a 2-D Gaussian (Figure 2H), the resulting residuals (Figure 2I) have large, nonrandom features with higher relative intensity values than the residuals in Figure 2E, indicating a poor quality of fit. This fit, which has an R2 value of 0.7, would not pass our quality-of-fit threshold, and thus this TAMRA centroid data point would be excluded from the final reconstructed image. The process described in Figure 2 for a single frame of one of our movies is repeated for all images acquired for this AuNR example. The calculated TAMRA centroid positions are shown as scatter plots in Figure 3A and B. A total of 43 788 frames were taken of this functionalized AuNR, and of those frames, 904 were found to be significant “on” events based on the fluorescence intensity algorithm, described in the Experimental Methods. Figure 3A shows the calculated TAMRA centroid positons of 837 TAMRA emission events that were successfully fit with R2 > 0.8 after the 3-dipole-modeled AuNR contribution has been subtracted off. In this case, more than 90% of the identified “on” events were successfully fit. On the other hand, Figure 3B shows that only 252 (28%) TAMRA emission events produce high-quality fits (R2 > 0.8) when the AuNR contribution is modeled as a 2D Gaussian, indicating that more than 70% of the identified “on” frames are excluded. Figure S4 shows the reconstructed images created when no quality-of-fit threshold is used. Importantly, all scatter plots for this example show a rod-like shape with the correct orientation, as determined by both the correlated AFM image (shown again in Figure 3G) and the calculated in-plane dipole angle that is generated by the 3dipole fit (ϕ = 48°, orange dashed line in Figure 3A and B and black dashed line in Figure 3G). However, due to the ability of the 3-dipole model to better approximate the AuNR dipolar emission, we have many more centroid data points in the reconstructed images, and we therefore gain a more complete map of the apparent locations of the dsDNA on the AuNR surface. Another way to visualize the apparent dsDNA locations is to create spatial frequency histograms, Figure 3C and D. The data from the scatter plots are binned in 4.4 nm × 4.4 nm bins (1/10th of an imaging pixel), and the centroid positions that fall within a particular bin are counted. The color map then represents the number of centroid positions in a particular bin and makes it easier to visually differentiate between areas that

Figure 3. (A) Scatter plot showing centroid positions of TAMRA emission (blue dots) and the average AuNR luminescence contribution (red ×) fit with the 3-dipole model. (B) Scatter plot showing centroid positions of TAMRA emission (blue dots) and the average AuNR luminescence contribution (red ×) fit with a 2-D Gaussian model. In (A) and (B), N is the number of TAMRA centroid positions that pass our goodness-of-fit criterion, and ϕ (orange line) is the in-plane orientation of the AuNR calculated from the 3-dipole model. (C and D) Spatial frequency histograms for the data shown in (A) and (B), respectively. The white × is the average (C) 3-dipolemodeled and (D) 2D Gaussian-modeled AuNR position. (E and F) Cumulative distribution plots of the x-values of centroid positions along the long axis of the AuNR in (A) and (B), respectively. The magenta ×’s and ◆ represent data points on opposite sides of the calculated average luminescence position (white × from C and D). (G) Correlated AFM image of the underlying AuNR, where the black line represents the same calculated ϕ as shown in (A) and (B). The out-of plane (θ) tilt angle for each AuNR is also included. (A), (B), (C), and (D) share a common 50 nm scale bar, and the scale bar in (G) is 50 nm.

have more centroid positions localized than others. Again, due to the larger number of centroid positions we were able to localize using the 3-dipole model as compared to the 2-D Gaussian model, we find a more robust reconstructed image of the underlying AuNR in Figure 3C than D. The white ×’s in the spatial frequency maps (and the red × in the scatter plots) represent the average AuNR luminescence position as determined by the different models. We find that the AuNR luminescence position is located in two distinctly different places between the two fitting approaches, consistent with our previous report.31 We assume that the average AuNR 19338

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

Article

The Journal of Physical Chemistry C

in Figure S6). Points are plotted as either black diamonds or magenta ×’s to show points that fall on either side of the AuNR center and highlight any asymmetry in the reconstructed AuNR length relative to the calculated luminescence center. We expect the CDF to plateau at its maximum value of 1 at a distance from center that corresponds to one-half the nanorod length (∼80/2 = 40 nm) plus the length of the DNA spacer (∼9.5 nm) for a total distance of 49.5 nm. From the CDF plot in Figure 3E, in which the 3-dipole model was used to fit the AuNR luminescence, the plateau happens ∼30 nm from the AuNR center, well below the expected distance of 49.5 nm. Thus, even when using this improved model for the AuNR luminescence, we still find reconstructed images with dimensions smaller than the expected size of the nanorod (the corresponding CDF plot for the y-values of the centroid points is shown in Figure S6D), consistent with our previous work.18 The data on either side of center (black vs magenta points) are symmetric due to the fact that the AuNR luminescence average position (as determined via the 3-dipole model) is centered with respect to the TAMRA centroid positions, as seen in Figure 3C. When the 2-D Gaussian model is used to model the AuNR luminescence, we see the data in the CDF plot (Figure 3F) are not symmetric due to the AuNR luminescence average position not being centered with respect to the TAMRA centroid position. For Figure 3F, the black data plateau at ∼15 nm and the magenta data plateau at ∼30 nm yielding a total reconstructed AuNR length of ∼45 nm (reconstructed width is ∼15 nm, CDF plot shown Figure S7). This is even smaller than the reconstructed dimensions calculated using the 3-dipole model. The dimensions of the AuNR from the AFM image, Figure 3G, were found to be ∼80 nm × ∼27 nm after accounting for tip effects that cause size overexaggeration. When the ∼9.5 nm dsDNA linker length is taken into account, the expected dimension for this functionalized AuNR is ∼99 nm × ∼46 nm. We discussed this size mismatch in previous work and hypothesized that a better AuNR model may help fix this issue; however, these data indicate that even with this improved model, we still reconstruct images well below the actual size of the nanorod.18 The size mismatch is also apparent in Figure 4 where the spatial frequency histogram maps from Figure 3C and D, and the corresponding AFM image, Figure 3G, are scaled to the

luminescence position should represent the center of the AuNR and would therefore be in the center of the TAMRA centroid positions if we are achieving uniform dsDNA coverage (unfortunately, a direct correlation between the calculated luminescence position and the center of the rod is impossible due to the AFM tip interacting with the beads used as alignment markers). While this is the case for the data fit with the 3-dipole model (Figure 3C), the luminescence is biased toward the upper left of the rod for the data calculated with the 2-D Gaussian (Figure 3D). We can understand this difference by analyzing the out-of plane tilt angle, θ, calculated using the 3-dipole model, which tells us that the AuNR is titled slightly out-of-plane at 87°. Previous work has shown that the accuracy of using a 2-D Gaussian to calculate the correct position from a dipole emitter will decrease if the emission being fit corresponds to an emitter that is tilted out-of-plane.31,48,49 In this case, the out-of-plane tilt of the rod leads to an incorrect location of the AuNR luminescence when calculated with a 2-D Gaussian, which is obvious from the data shown in Figure 3D. For comparison, we also directly subtracted the nearest raw AuNR luminescence image data (IAuNR) from the combined “on” frames to obtain the TAMRA contribution, which was then fit to eq 1. This strategy removes the dependence on the AuNR luminescence model and has been used by several other research groups to remove the contribution of Au luminescence backgrounds when fitting fluorescence from untethered, diffusing emitters.24,50 The analysis involves finding the nearest identified “off” frames both before and after each “on” frame, averaging the two “off” frames, and then subtracting the result from the “on” frame and fitting the remainder to eq 1. The resulting spatial frequency histogram maps from using this technique are included in Figure S5. The examples shown in Figure S5 qualitatively appear similar to the reconstructed images created using the 3-dipole modeled AuNR luminescence contribution. However, for the majority of the functionalized AuNRs, the number of centroid positions (N) that pass the quality-of-fit threshold when using this process is lower than the number of centroids calculated when using the 3-dipole model to account for the AuNR luminescence. We attribute this to the fact that the direct subtraction approach cannot account for subpixel drift that occurs between nearby frames. We are particularly sensitive to this drift due to the fact that we impose strict criteria for assigning a frame as “off” (e.g., Au luminescence only), and thus the temporal difference between a subtracted “on” and “off” frame can be many frames long. Further, using this raw subtraction approach does not provide any information about the AuNR orientation and luminescence position, requiring us to fit the AuNR luminescence to the 3dipole model to extract these data. Because we need to fit the AuNR anyway and we find that we get more high-quality fits (R2 > 0.8) using the 3-dipole subtraction approach, we believe that this strategy is superior for analyzing data when fluorophores are tethered to the nanoparticle surface, as presented here. To compare the calculated size of the reconstructed images to the actual size of the AuNR, we calculated cumulative distribution function (CDF) plots, as shown in Figure 3E and F. Briefly, the data in the centroid position scatter plots (Figure 3A and B) are rotated such that the long axis of the AuNR lies on the x-axis (Figure S6). We then calculate the fraction of centroid positions that fall between the average luminescence position (which we designate as the center of the AuNR) and a particular distance from that center (more details can be found

Figure 4. Reconstructed images from Figure 3C (left) and D (right) and the AFM image from Figure 3G were scaled to an equal size and overlaid to show the size mismatch between expected and reconstructed dimensions. The gray outline represents the expected functionalized AuNR dimensions, after taking into account AFM tip effects. 19339

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

Article

The Journal of Physical Chemistry C same scale bar and are qualitatively overlaid upon each other. The reconstructed images are clearly smaller than the AFM image, even when we take into account the size overexaggeration caused by tip effects from the AFM (the corrected size is marked with the gray line). The width of the reconstructed functionalized AuNR image, when using the 3dipole model, provides a closer agreement to the expected dimension than when using the 2-D Gaussian; however, the length is undersized in both cases. Figure 5 shows a montage of different AuNRs illustrating the effects of fitting the AuNR contribution with the 3-dipole model (left column) or the 2-D Gaussian model (middle column) on the calculated spatial frequency histograms. The left column, Figure 5A, D, G, and J, shows spatial frequency histogram maps that have been constructed from “on” events in which the 3-dipole-modeled AuNR luminescence contribution has been subtracted away, leaving only TAMRA emission. The second column, Figure 5B, E, H, and K, shows spatial frequency maps that have been constructed using 2-D Gaussian-modeled AuNR luminescence to subtract the AuNR contribution from the TAMRA emission. The histogram maps in each row correspond to the same functionalized AuNR. The number of TAMRA emission centroid positions that were used to create the histogram maps is included on each map. More centroid points pass our quality-of-fit screening when the 3-dipole model was used than the 2-D Gaussian model, consistent with the trend from Figure 3. The correlated AFM images for each AuNR are in the right column, Figure 5C, F, I, and L. The white × in each spatial frequency histogram marks the average AuNR position calculated with the different models. As with Figure 3, we see a larger offset between the luminescence positions calculated with the two different models as the out-ofplane tilt increases (decreasing θ). Figure 5 once again shows that using an improved fitting strategy for modeling the AuNR luminescence contribution does not address the size mismatch between the reconstructed image size and the actual AuNR size. Thus, the central hypothesis of this Article, that an improved fit model for the AuNR luminescence could generate accurate reconstructed image sizes of dsDNA-labeled AuNRs, can be rejected. As an added feature of this analysis, we also observe that the spatial frequency histograms in Figure 5 reveal diversity in the locations of the TAMRA activity, suggesting either local heterogeneity in the binding of the TAMRA-labeled dsDNA or locally enhanced/suppressed TAMRA activity on the AuNR surface. For example, the top row of Figure 5A,B shows more activity on the ends of the nanorod, with a dearth of activity near the center; this behavior is also apparent in the second row. In the second row, there are more centroid localization events in the upper right portion of the nanorod than the lower left, suggesting either more TAMRA-labeled dsDNA at this site or more active TAMRA emitters. This trend is also apparent in the other examples, particularly the data shown in the third row, where the upper portion of the AuNR shows much more TAMRA activity than the lower portion. Similar behavior has been previously observed by us.18 What is of particular note here is that this apparent asymmetry is present regardless of the model used to fit the AuNR data, and that there is excellent qualitative agreement between TAMRA activity observed in the spatial intensity maps using the 3-dipole fit model (Figure 5, left column), the 2-D Gaussian model (Figure 5, center column), and even the raw data subtraction approach (Figure S5). This suggests that the asymmetry is not an artifact of the

Figure 5. Left column (A, D, G, and J): Spatial frequency histograms showing TAMRA emission for four different AuNRs. The AuNR luminescence was fit with the 3-dipole model, and the average location is shown as a white ×. Middle column (B, E, H, and K): Spatial frequency histograms for the same four AuNRs, with the AuNR luminescence fit with a 2-D Gaussian. The white × represents the average AuNR luminescence position calculated with the 2-D Gaussian model. Right column (C, F, I, and L): Correlated AFM images of the underlying AuNRs. The black line represents the calculated ϕ from the 3-dipole model. The out-of plane (θ) tilt angle for each AuNR is also included. Each row corresponds to the same TAMRA-labeled AuNR. N equals the number of TAMRA centroid positions that were localized. All of the spatial frequency histograms share a common 50 nm scale bar, and the AFM images have a 50 nm scale bar.

model, but is instead a real experimental observation most likely associated with local heterogeneity in the binding/activity of the TAMRA-labeled dsDNA. Figure 6 shows the CDF plots for both the length and the width of all of the AuNRs discussed in this Article (a color legend is included in the figure caption), and once again, we find that the size of the reconstructed AuNR is well below the expected dimensions, regardless of the model used. Figure 6A and B shows the length and width CDF plots that were created from the TAMRA centroid data in which the 3-dipole-modeled AuNR luminescence was subtracted, while Figure 6C and D 19340

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

Article

The Journal of Physical Chemistry C

nm, which means the reconstructed length should be ∼108 nm when dsDNA linker length is taken into account. Ideally, if we were reconstructing the proper length, we would expect the plateau to happen around ∼54 nm for data points on both sides of zero if the lines were symmetric, yet we observe that 100% of the points are accounted for at ∼30 nm from center for all examples shown, regardless of model. This same trend is also evident in the width CDF plots, Figure 6B and D. The expected width for the AuNRs used is 30 ± 2 nm, so the reconstructed images should be ∼49 nm wide when the dsDNA linker is taken into account. This means the data points should not plateau until ∼25 nm for data on both sides of zero, assuming the lines are symmetric. In the CDF plots using either the 3dipole (Figure 6B) or the 2-D Gaussian (Figure 6D) models, the widths are still below the expected value, although we find that the 3-dipole model yields larger overall widths than the 2D Gaussian, suggesting at least some improvement in the reconstructed image size when using this fit. Finally, we compile the results of our studies into Table 1, which compares the results for the five rods discussed in this study. In addition to the orientation data provided by the 3dipole fit (ϕ and θ), we include the scattering maximum of each AuNR (λscat, max, from Figure S1), and the approximate dimensions as determined by AFM and our super-resolution fits. For the AFM data, we compare the difference between the AuNR height and width and attribute the difference to tip effects (dtip = 1/2(width − height)). Assuming that the rod is symmetric, we use the height as an indicator of the nanorod width, and then calculate the length by subtracting off dtip*2 from the measured length. The approximate length and width values for the super-resolution fits using the different models are determined by visual inspection of the plateau region of the CDF plots (Figure S9). While this is not the most rigorous method for dimension determination, we found that the data could not be well-modeled using traditional probability functions, and we did not want to use the most extreme centroid positions to calculate the dimensions. Thus, these values should be taken as rough approximations, with error bars of ±5 nm. Importantly, even if we vastly overestimated the dimensions from the CDF plots, we would never match the measured dimensions of the AuNR, indicating that the choice of fitting model is not the origin of the size mismatch between the reconstructed images and the AFM data. However, we do note strong agreement between the aspect ratio (length/width) calculated from the CDF plots and the aspect ratio calculated from the AFM data (accounting for the dsDNA spacer length), indicating that the super-resolution fits are yielding a physically

Figure 6. Cumulative distribution function plots depicting centroid positions along the long (A) and short (B) axes of the AuNR, with luminescence fit to the 3-dipole model. Cumulative distribution function plots depicting centroid positions along the long (C) and short (D) axis of the AuNR, with luminescence fit to a 2-D Gaussian model. Within a given cumulative distribution plot, each color corresponds to a different AuNR. Among all four plots, each color corresponds to the same functionalized AuNR example. The ×’s and diamonds represent data points on opposite sides of the calculated average luminescence position. Color legend: magenta corresponds to the data used to reconstructed images in Figure 3A−D, blue corresponds to Figure 5A,B, green corresponds to Figure 5D,E, orange corresponds to Figure 5G,H, and black corresponds to Figure 5J,K.

shows the length and width CDF plots when the AuNR was modeled as a 2-D Gaussian. We chose to use CDF plots because we can combine the CDF plots from all AuNRs in a single plot for easy viewing and comparison (the CDF plots for individual AuNRs are shown in Figure S9). The overall trend in the length CDF plots, Figure 6A and C, is that the data points plateau sooner than the expected distance from center. The average length of the AuNRs discussed in this Article is 89 ± 8

Table 1. Tabulated Results for the Functionalized AuNRs in This Article

Figure 3 Figure 5A−C Figure 5D−F Figure 5G−I Figure 5J−L a

ϕ (deg)a

θ (deg)a

λscat,max (nm)

approx length × width (nm), from AFM (AR)b

approx length × width (nm), adding DNA linker (AR)b

approx length × width (nm), from 3-dipole CDF plots (AR)b

approx length × width (nm), from 2-D Gaussian CDF plots (AR)b

48 348

87 89

∼690 ∼680

∼80 × ∼27 (3.0) ∼91 × ∼29 (3.1)

∼99 × ∼46 (2.1) ∼110 × ∼48 (2.3)

∼60 × ∼50 (1.2) ∼60 × ∼35 (1.7)

∼45 × ∼15 (3) ∼60 × ∼30 (2)

134

88

∼640

∼95 × ∼33 (2.9)

∼114 × ∼52 (2.2)

∼70 × ∼30 (2.3)

∼70 × ∼20 (3.5)

91

87

∼675

∼95 × ∼29 (3.3)

∼114 × ∼48 (2.3)

∼55 × ∼25 (2.2)

∼55 × ∼25 (2.2)

37

87

∼640

∼81 × ∼31 (2.6)

∼100 × ∼50 (2)

∼60 × ∼30 (2)

∼50 × ∼10 (2)

Calculated from the 3-dipole model. bAR = aspect ratio, defined as length divided by width. 19341

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

The Journal of Physical Chemistry C

■

relevant structure parameter in addition to the nanorod shape and orientation.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected].

CONCLUSIONS In this Article, we compared the effects of different AuNR luminescence fit models on the reconstructed images of fluorescently labeled dsDNA bound to the surface of AuNRs. By using the 3-dipole model, we were able to pass more TAMRA emission events through our fitting threshold to get a more robust map of the apparent locations of bound dsDNA in the reconstructed images, and the model provides useful output parameters such as θ and ϕ for the underlying AuNR. Unfortunately, subtracting off a better modeled AuNR contribution did not fix the size mismatch issues between the expected AuNR dimensions and the dimensions of the functionalized AuNR in the super-resolution reconstructed images. However, we do find that the reconstructed images yield aspect ratios that are consistent with the aspect ratio of the labeled AuNRs. We believe that other factors are contributing to this size-mismatch issue, such as altered TAMRA photophysics caused by plasmonic interactions with the AuNR that can change the triplet state lifetime of TAMRA and/or induce TAMRA−AuNR coupling26 or the formation of image dipoles within the AuNR that can perturb the output point spread function.51 Both factors will impact the reconstructed images of the labeled AuNRs by causing smaller-than-expected images; however, more experiments are needed to fully understand the magnitude of their impacts. Despite the smaller-than-expected reconstructed images, we were able to accurately model AuNR luminescence and TAMRA emission within a diffraction-limited spot to map the position of the AuNR and gain insight into the apparent heterogeneity of dsDNA on the AuNR surface. With both models, we observed similar asymmetry in terms of apparent dsDNA binding locations in the spatial frequency histograms, suggesting that the observed binding heterogeneity is real and not a fit artifact. Finally, we showed that the luminescence signal can serve as a benchmark to combine multiple data sets together; therefore, data collection is only limited to how long TAMRA molecules continue to undergo photoswitching. This increases the likelihood of probing all of the dsDNA that are attached to the AuNR surface. By fitting the luminescence to a 3-dipole model, we extract multiple parameters associated with the AuNR orientation, which will ultimately negate the need for correlated AFM imaging to confirm that our super-resolution image correctly maps the underlying shape of the AuNR support.

■

Article

Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS The work presented in this Article was supported by the National Science Foundation under Grant No. CBET-1402610 and Welch Foundation Award No. F-1699.

■

REFERENCES

(1) Murphy, C. J.; Gole, A. M.; Stone, J. W.; Sisco, P. N.; Alkilany, A. M.; Goldsmith, E. C.; Baxter, S. C. Gold Nanoparticles in Biology: Beyond Toxicity to Cellular Imaging. Acc. Chem. Res. 2008, 41, 1721− 1730. (2) Jain, P. K.; Huang, X.; El-Sayed, I. H.; El-Sayed, M. A. Noble Metals on the Nanoscale: Optical and Photothermal Properties and Some Applications in Imaging, Sensing, Biology, and Medicine. Acc. Chem. Res. 2008, 41, 1578−1586. (3) Bardhan, R.; Lal, S.; Joshi, A.; Halas, N. J. Theranostic Nanoshells: From Probe Design to Imaging and Treatment of Cancer. Acc. Chem. Res. 2011, 44, 936−946. (4) West, J. L.; Halas, N. J. Engineered Nanomaterials for Biophotonics Applications: Improving Sensing, Imaging, and Therapeutics. Annu. Rev. Biomed. Eng. 2003, 5, 285−292. (5) Huang, X.; El-Sayed, I. H.; Qian, W.; El-Sayed, M. A. Cancer Cell Imaging and Photothermal Therapy in the Near-Infrared Region by Using Gold Nanorods. J. Am. Chem. Soc. 2006, 128, 2115−2120. (6) Wijaya, A.; Schaffer, S. B.; Pallares, I. G.; Hamad-Schifferli, K. Selective Release of Multiple DNA Oligonucleotides from Gold Nanorods. ACS Nano 2009, 3, 80−86. (7) Huschka, R.; Zuloaga, J.; Knight, M. W.; Brown, L. V.; Nordlander, P.; Halas, N. J. Light-Induced Release of DNA from Gold Nanoparticles: Nanoshells and Nanorods. J. Am. Chem. Soc. 2011, 133, 12247−12255. (8) Mackey, M. A.; Ali, M. R. K.; Austin, L. A.; Near, R. D.; El-Sayed, M. A. The Most Effective Gold Nanorod Size for Plasmonic Photothermal Therapy: Theory and In Vitro Experiments. J. Phys. Chem. B 2014, 118, 1319−1326. (9) Huang, X.; Jain, P. K.; El-Sayed, I. H.; El-Sayed, M. A. Plasmonic Photothermal Therapy (PPTT) Using Gold Nanoparticles. Lasers Med. Sci. 2008, 23, 217−228. (10) Lee, A.; Andrade, G. F. S.; Ahmed, A.; Souza, M. L.; Coombs, N.; Tumarkin, E.; Liu, K.; Gordon, R.; Brolo, A. G.; Kumacheva, E. Probing Dynamic Generation of Hot-Spots in Self-Assembled Chains of Gold Nanorods by Surface-Enhanced Raman Scattering. J. Am. Chem. Soc. 2011, 133, 7563−7570. (11) Fu, Y.; Zhang, J.; Lakowicz, J. R. Plasmon-Enhanced Fluorescence from Single Fluorophores End-Linked to Gold Nanorods. J. Am. Chem. Soc. 2010, 132, 5540−5541. (12) Alkilany, A. M.; Lohse, S. E.; Murphy, C. J. The Gold Standard: Gold Nanoparticle Libraries To Understand the Nano−Bio Interface. Acc. Chem. Res. 2013, 46, 650−661. (13) Murphy, C. J.; Sau, T. K.; Gole, A. M.; Orendorff, C. J.; Gao, J.; Gou, L.; Hunyadi, S. E.; Li, T. Anisotropic Metal Nanoparticles: Synthesis, Assembly, and Optical Applications. J. Phys. Chem. B 2005, 109, 13857−13870. (14) Yorulmaz, M.; Khatua, S.; Zijlstra, P.; Gaiduk, A.; Orrit, M. Luminescence Quantum Yield of Single Gold Nanorods. Nano Lett. 2012, 12, 4385−4391. (15) Fang, Y.; Chang, W.-S.; Willingham, B.; Swanglap, P.; Dominguez-Medina, S.; Link, S. Plasmon Emission Quantum Yield of Single Gold Nanorods as a Function of Aspect Ratio. ACS Nano 2012, 6, 7177−7184.

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jpcc.5b04993. Dark field microscopy scatter spectra for the functionalized AuNR discussed, diagram describing the in-plane and out-of-plane angles, original images shown in Figure 1 (not contrasted to show side lobes), reconstructed images created without using a quality-of-fit threshold, reconstructed images created subtracting adjacent “off” frames from “on frames”, instructions for constructing CDF plots, the associated scatter plots for the functionalized AuNRs shown in Figure 5, and individual CDF plots for all of the functionalized AuNRs (PDF) 19342

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343

Article

The Journal of Physical Chemistry C (16) Link, S.; El-Sayed, M. A. Shape and Size Dependence of Radiative, Non-Radiative and Photothermal Properties of Gold Nanocrystals. Int. Rev. Phys. Chem. 2000, 19, 409−453. (17) Blythe, K. L.; Mayer, K. M.; Weber, M. L.; Willets, K. A. Ground State Depletion Microscopy for Imaging Interactions Between Gold Nanowires and Fluorophore-Labeled Ligands. Phys. Chem. Chem. Phys. 2013, 15, 4136−4145. (18) Blythe, K. L.; Titus, E. J.; Willets, K. A. Triplet-State-Mediated Super-Resolution Imaging of Fluorophore-Labeled Gold Nanorods. ChemPhysChem 2014, 15, 784−793. (19) Su, L.; Lu, G.; Kenens, B.; Rocha, S.; Fron, E.; Yuan, H.; Chen, C.; Van Dorpe, P.; Roeffaers, M. B. J.; Mizuno, H.; et al. Visualization of Molecular Fluorescence Point Spread Functions via Remote Excitation Switching Fluorescence Microscopy. Nat. Commun. 2015, 6, 6287. (20) Lin, H.; Centeno, S. P.; Su, L.; Kenens, B.; Rocha, S.; Sliwa, M.; Hofkens, J.; Uji-i, H. Mapping of Surface-Enhanced Fluorescence on Metal Nanoparticles Using Super-Resolution Photoactivation Localization Microscopy. ChemPhysChem 2012, 13, 973−981. (21) Bates, M.; Huang, B.; Dempsey, G. T.; Zhuang, X. Multicolor Super-Resolution Imaging with Photo-Switchable Fluorescent Probes. Science 2007, 317, 1749−1753. (22) Xiao, L.; Qiao, Y.; He, Y.; Yeung, E. S. Three Dimensional Orientational Imaging of Nanoparticles with Darkfield Microscopy. Anal. Chem. 2010, 82, 5268−5274. (23) Fölling, J.; Bossi, M.; Bock, H.; Medda, R.; Wurm, C. A.; Hein, B.; Jakobs, S.; Eggeling, C.; Hell, S. W. Fluorescence Nanoscopy by Ground-State Depletion and Single-Molecule Return. Nat. Methods 2008, 5, 943−945. (24) Zhou, X.; Andoy, N. M.; Liu, G.; Choudhary, E.; Han, K.-S.; Shen, H.; Chen, P. Quantitative Super-Resolution Imaging Uncovers Reactivity Patterns on Single Nanocatalysts. Nat. Nanotechnol. 2012, 7, 237−241. (25) Tam, F.; Goodrich, G. P.; Johnson, B. R.; Halas, N. J. Plasmonic Enhancement of Molecular Fluorescence. Nano Lett. 2007, 7, 496− 501. (26) Wertz, E.; Isaacoff, B. P.; Flynn, J. D.; Biteen, J. S. SingleMolecule Super-Resolution Microscopy Reveals How Light Couples to a Plasmonic Nanoantenna on the Nanometer Scale. Nano Lett. 2015, 15, 2662−2670. (27) Steinhauer, C.; Forthmann, C.; Vogelsang, J.; Tinnefeld, P. Superresolution Microscopy on the Basis of Engineered Dark States. J. Am. Chem. Soc. 2008, 130, 16840−16841. (28) Jones, M. R.; Macfarlane, R. J.; Lee, B.; Zhang, J.; Young, K. L.; Senesi, A. J.; Mirkin, C. A. DNA-Nanoparticle Superlattices Formed from Anisotropic Building Blocks. Nat. Mater. 2010, 9, 913−917. (29) Hill, H. D.; Millstone, J. E.; Banholzer, M. J.; Mirkin, C. A. The Role Radius of Curvature Plays in Thiolated Oligonucleotide Loading on Gold Nanoparticles. ACS Nano 2009, 3, 418−424. (30) Lin, Q.-Y.; Li, Z.; Brown, K. A.; O’Brien, M. N.; Ross, M. B.; Zhou, Y.; Butun, S.; Chen, P.-C.; Schatz, G. C.; Dravid, V. P.; et al. Strong Coupling Between Plasmonic Gap Modes and Photonic Lattice Modes in DNA-Assembled Gold Nanocube Arrays. Nano Lett. 2015, 15, 4699−4703. (31) Titus, E. J.; Willets, K. A. Accuracy of Superlocalization Imaging Using Gaussian and Dipole Emission Point-Spread Functions for Modeling Gold Nanorod Luminescence. ACS Nano 2013, 7, 6258− 6267. (32) Rust, M. J.; Bates, M.; Zhuang, X. Sub-Diffraction-Limit Imaging by Stochastic Optical Reconstruction Microscopy (STORM). Nat. Methods 2006, 3, 793−796. (33) Hess, S. T.; Girirajan, T. P. K.; Mason, M. D. Ultra-High Resolution Imaging by Fluorescence Photoactivation Localization Microscopy. Biophys. J. 2006, 91, 4258−4272. (34) Blythe, K. L.; Titus, E. J.; Willets, K. A. Objective-Induced Point Spread Function Aberrations and Their Impact on Super-Resolution Microscopy. Anal. Chem. 2015, 87, 6419−6424. (35) Mayer, K. M.; Lee, S.; Liao, H.; Rostro, B. C.; Fuentes, A.; Scully, P. T.; Nehl, C. L.; Hafner, J. H. A Label-Free Immunoassay

Based Upon Localized Surface Plasmon Resonance of Gold Nanorods. ACS Nano 2008, 2, 687−692. (36) Ming, T.; Zhao, L.; Yang, Z.; Chen, H.; Sun, L.; Wang, J.; Yan, C. Strong Polarization Dependence of Plasmon-Enhanced Fluorescence on Single Gold Nanorods. Nano Lett. 2009, 9, 3896−3903. (37) Liao, H.; Hafner, J. H. Gold Nanorod Bioconjugates. Chem. Mater. 2005, 17, 4636−4641. (38) Weber, M. L.; Willets, K. A. Correlated Super-Resolution Optical and Structural Studies of Surface-Enhanced Raman Scattering Hot Spots in Silver Colloid Aggregates. J. Phys. Chem. Lett. 2011, 2, 1766−1770. (39) Stranahan, S. M.; Titus, E. J.; Willets, K. A. SERS Orientational Imaging of Silver Nanoparticle Dimers. J. Phys. Chem. Lett. 2011, 2, 2711−2715. (40) Stranahan, S. M.; Willets, K. A. Super-Resolution Optical Imaging of Single-Molecule SERS Hot Spots. Nano Lett. 2010, 10, 3777−3784. (41) Shegai, T.; Brian, B.; Miljković, V. D.; Käll, M. Angular Distribution of Surface-Enhanced Raman Scattering from Individual Au Nanoparticle Aggregates. ACS Nano 2011, 5, 2036−2041. (42) Shegai, T.; Li, Z.; Dadosh, T.; Zhang, Z.; Xu, H.; Haran, G. Managing Light Polarization via Plasmon−molecule Interactions Within an Asymmetric Metal Nanoparticle Trimer. Proc. Natl. Acad. Sci. U. S. A. 2008, 105, 16448−16453. (43) Lieb, M. A.; Zavislan, J. M.; Novotny, L. Single-Molecule Orientations Determined by Direct Emission Pattern Imaging. J. Opt. Soc. Am. B 2004, 21, 1210−1215. (44) Huang, C.; Bouhelier, A.; Colas des Francs, G.; Bruyant, A.; Guenot, A.; Finot, E.; Weeber, J.-C.; Dereux, A. Gain, Detuning, and Radiation Patterns of Nanoparticle Optical Antennas. Phys. Rev. B: Condens. Matter Mater. Phys. 2008, 78, 155407. (45) Bartko, A. P.; Dickson, R. M. Imaging Three-Dimensional Single Molecule Orientations. J. Phys. Chem. B 1999, 103, 11237−11241. (46) Böhmer, M.; Enderlein, J. Orientation Imaging of Single Molecules by Wide-Field Epifluorescence Microscopy. J. Opt. Soc. Am. B 2003, 20, 554−559. (47) Sick, B.; Hecht, B.; Novotny, L. Orientational Imaging of Single Molecules by Annular Illumination. Phys. Rev. Lett. 2000, 85, 4482− 4485. (48) Enderlein, J.; Toprak, E.; Selvin, P. R. Polarization Effect on Position Accuracy of Fluorophore Localization. Opt. Express 2006, 14, 8111−8120. (49) Engelhardt, J.; Keller, J.; Hoyer, P.; Reuss, M.; Staudt, T.; Hell, S. W. Molecular Orientation Affects Localization Accuracy in Superresolution Far-Field Fluorescence Microscopy. Nano Lett. 2011, 11, 209−213. (50) Donehue, J. E.; Wertz, E.; Talicska, C. N.; Biteen, J. S. PlasmonEnhanced Brightness and Photostability from Single Fluorescent Proteins Coupled to Gold Nanorods. J. Phys. Chem. C 2014, 118, 15027−15035. (51) Ropp, C.; Cummins, Z.; Nah, S.; Fourkas, J. T.; Shapiro, B.; Waks, E. Nanoscale Probing of Image-Dipole Interactions in a Metallic Nanostructure. Nat. Commun. 2015, 6, 6558.

19343

DOI: 10.1021/acs.jpcc.5b04993 J. Phys. Chem. C 2015, 119, 19333−19343